New simulation methods for the prediction of binding free energies

Abstract

The calculation of free energies of binding is arguably one of the most important challenges for the molecular modeller. However, established methods for these calculations are too slow to be useful for practical drug design problems. A host of methods have been proposed for increasing the efficiency of such calculations. The work presented in this thesis firstly studies the Linear Interaction Energy method. The method is applied to two protein-ligand systems (neuraminidase and OppA) and to the calculation of free energies of hydration. The protein systems were chosen because they were considered to be particularly challenging to the methodology. The method was approached in a very different way from previous applications which have simply calculated the best fit to a given data set using a predefined series of variables. In contrast the approach proposed in this thesis is to use the information content of the data in conjunction with carefully applied statistical techniques to assess the importance of a wide range of variables for inclusion in a predictive equation. Application of the method to the protein systems revealed quite categorically that there is no one LIE equation which can be predictive for all systems. Furthermore, it seems clear that the variables present in predictive models are system dependent. It was also possible to rationalise the variables selected on the basis of the experimental data.

The second approach to the fast free energy problem was to attempt to combine the Generalised Born/Surface Area (GB/SA) model with the LIE method. The technique was applied to the same neuraminidase system used in the original LIE work. The results showed a significant increase in sampling using GB/SA. Unfortunately, owing to the inability of the continuum solvation model to represent specific solute-solvent interactions, the resulting sampling was not considered to be representative of the real system. Attempts to modify the LIE method for the new energy terms identified the same variables to be present in the most predictive equation as for the explicit simulations. However since these terms no longer contained solute-solvent information, the predictive ability was dramatically reduced.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata